# Kuanyin First-Line Moderation

`@v2/kuanyin-first-line-moderation` is the V2 adapter that replaces Two Hat,
Community Sift, Hive AI, and AWS Rekognition as first-line moderation. V2 text,
report, replay transcript, player-name, and UGC caption checks now route through
`@kuanyin/precognition` plus the `@kuanyin/foundation` harm taxonomy.

## Primary Path

- `@kuanyin/precognition` classifies intent, harm potential, target signals,
  urgency, and coded language.
- `@kuanyin/foundation` supplies harm keyword patterns, severity thresholds,
  emotional assessment, and intervention decisions.
- The adapter emits V2 decisions for allow, friction, hold-for-review, block,
  and emergency escalation.

## Legacy Fallback

Two Hat and Community Sift are retired for V2 and cannot be used as text
fallbacks. Hive AI and AWS Rekognition are retained only as media fallback
providers for image or video coverage gaps, such as a UGC image with no
extractable text signal.

If Kuanyin text classification is unavailable, V2 routes the content to manual
review instead of falling back to legacy text vendors.

## Rollback Boundary

Moderation is off rollback. The adapter can hold, block, escalate, or add
friction to publishing and chat/report flows, but it cannot affect deterministic
match simulation, rollback inputs, or competitive frame outcomes.

## Audit Publication

Moderation decisions publish a canonical `v2.moderation.decision_published`
event through `@oshun/audit-platform`. V2 publishes the decision context; Oshun
retains and exports the shared moderation audit record.